Video-based Gait Recognition

In ambient intelligence, the combination of human motion analysis and biometrics has been a prevalent research field. Gait recognition, the second biometrics identification technology based on motion vision, is used to signify the identity of individuals in image sequences'by the way they walk'. Compared with other biometrics, such as fingerprint, face and iris, gait is unobtrusive and does not require to contact, so gait recognition may be performed at a distance surreptitiously. This thesis, in both the theoretical and the practical perspective, probes into gait recognition with the videos as input. The main contributions of this thesis are summarized as follows:1) Proposed a method using amplitude spectrum and reflective symmetry of key frames as gait features to recognize test sequence. Amplitude spectrum (Fourier spectrum) of frequency domain and psychological studies which suggested that gait is a symmetrical pattern of motion are applied to analyse gait. Amplitude spectrum describes the frequency feature of gait image and reflective symmetry implicitly represents the habit with which person's leg and body swings. Experiments on USF dataset show that recognition rate using symmetry feature with frequency feature and proportion of body is higher than only exploiting the frequency feature and the proportion. Reflective symmetry is not a unique but a useful feature and it can be easily combined with other features to improve the recognition rate. Proposed algorithm is simple to implement and insensitive to the changes of walking speed, background, and so on.2) Developed new wavelet velocity moments, wavelet reflective symmetry moments and combined wavelet velocity moments and reflective symmetry features together to describe and recognize gait. Wavelet moments are new moment features which combine the characteristics of wavelet and moment. So they have the features of moments: translation, scale and rotation invariance, and the trait of wavelet analysis (strongly anti-noise performance and local analysis with multi-resolution). Wavelet moments avoid the computation of high order geometric moments and, at the same time can strengthen the analysis ability to subtle image feature, thus showing good performance in classifying similar image objects with subtle differences. Because people are walking, velocity is an important feature to be used to discriminate them, reflective symmetry implicitly represents the habit with which person's leg and body swings. Wavelet velocity moments, wavelet reflective symmetry moments and combined features are...